Author: Robert Hyatt
Date: 08:30:26 05/07/04
Go up one level in this thread
On May 07, 2004 at 04:16:12, martin fierz wrote:
>On May 06, 2004 at 21:44:30, Robert Hyatt wrote:
>
>>I re-read your stuff. I think the way you did the calculation was a bit more
>>favorable than the way I do it. IE I always use the _last_ PV produced by the
>>1-cpu search, and the time for that _same_ PV on the N-cpu search.
>
>and what do you do if that same PV of yours doesn't turn up again?? that's the
>reason i did it this way - there were two instances where after completion of
>the ply the 4-way crafty wanted to play a different move than 1-way crafty.
>
It always does. I am talking about (say) something like this:
1cpu run:
11 32.01 -0.34 1. ... Nce7 2. Qxf4 fxg2 3. Rf2 Bf5
4. Nb5 Bh6 5. Qxc7 Bxc1 6. Rxc1 Nh6
7. Rxg2 Qxc7 8. Nxc7+
11-> 49.35 -0.34 1. ... Nce7 2. Qxf4 fxg2 3. Rf2 Bf5
4. Nb5 Bh6 5. Qxc7 Bxc1 6. Rxc1 Nh6
7. Rxg2 Qxc7 8. Nxc7+
12 1:15 0.03 1. ... Nce7 2. Qxf4 fxg2 3. Rf2 Bf5
4. Bb5+ c6 5. dxc6 bxc6 6. Bxc6+ Nxc6
7. Qxf5 Qd7 8. Qxd7+ Kxd7 9. d5 Bc5
10. dxc6+ Kxc6 11. Kxg2 Bxf2 12. Kxf2
12 4:44 -0.32 1. ... Na5 2. Bb5+ c6 3. Qxf4 cxb5
4. Nxb5 Nf6 5. Nc7+ Kd7 6. Nxa8 fxg2
7. Kxg2 Rg8 8. Kh1 Bh3 9. Qxf6 Bxf1
10. Qxf1 Qxa8 11. Qxf7+
time=5:00 cpu=99% mat=-3 n=570081156 fh=92% nps=1.90M
4cpu run:
10-> 7.10 -0.17 1. ... Nce7 2. Qxf4 fxg2 3. Qxf7+ Kd7
4. Kxg2 Qe8 5. h3 Qxf7 6. Bb5+ c6 7.
dxc6+ bxc6 8. Rxf7 Bxh3+ 9. Kxh3 cxb5
10. Nxb5 (s=3)
11 11.14 -0.34 1. ... Nce7 2. Qxf4 fxg2 3. Rf2 Bf5
4. Nb5 Bh6 5. Qxc7 Bxc1 6. Rxc1 Nh6
7. Rxg2 Qxc7 8. Nxc7+ (s=2)
11-> 16.98 -0.34 1. ... Nce7 2. Qxf4 fxg2 3. Rf2 Bf5
4. Nb5 Bh6 5. Qxc7 Bxc1 6. Rxc1 Nh6
7. Rxg2 Qxc7 8. Nxc7+
12 25.90 0.03 1. ... Nce7 2. Qxf4 fxg2 3. Rf2 Bf5
4. Bb5+ c6 5. dxc6 bxc6 6. Bxc6+ Nxc6
7. Qxf5 Qd7 8. Qxd7+ Kxd7 9. d5 Bc5
10. dxc6+ Kxc6 11. Kxg2 Bxf2 12. Kxf2
12 1:41 -0.32 1. ... Na5 2. Bb5+ c6 3. Qxf4 cxb5
4. Nxb5 Nf6 5. Nc7+ Kd7 6. Nxa8 fxg2
7. Kxg2 Rg8 8. Kh1 Bh3 9. Qxf6 Bxf1
10. Qxf1 Qxa8 11. Qxf7+
12-> 1:48 -0.32 1. ... Na5 2. Bb5+ c6 3. Qxf4 cxb5
4. Nxb5 Nf6 5. Nc7+ Kd7 6. Nxa8 fxg2
7. Kxg2 Rg8 8. Kh1 Bh3 9. Qxf6 Bxf1
10. Qxf1 Qxa8 11. Qxf7+
13 2:53 -0.03 1. ... Na5 2. Bb5+ c6 3. Qxf4 cxb5
4. Qe5+ Ne7 5. Ne4 Qb6 6. Nf6+ Kd8
7. Nxg4 f6 8. Nxf6 Nc4 9. Qf4 fxg2
10. Kxg2
13-> 4:06 -0.03 1. ... Na5 2. Bb5+ c6 3. Qxf4 cxb5
4. Qe5+ Ne7 5. Ne4 Qb6 6. Nf6+ Kd8
7. Nxg4 f6 8. Nxf6 Nc4 9. Qf4 fxg2
10. Kxg2
time=5:00 cpu=397% mat=-3 n=2214294718 fh=92% nps=7.38M
I use 4:44 as the 1cpu time, 1:41 as the 4cpu time...
>i wanted to take the last known PV, but couldn't because of this.
>
>cheers
> martin
>
>>The first
>>move is more difficult to search in parallel than the remainder of the ply-1
>>moves... that is why I split at the root when possible in fact.
>>
>>But, as the saying goes, you can prove nearly anything depending on how you
>>choose to interpret data. Your interpretation is hardly invalid... And might
>>sometimes be more favorable rather than less.
>>
>>Who knows...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.